DISCUSSION PAPER SERIES

IZA DP No. 12122

Jenny WilliamsDon Weatherburn

Can Electronic Monitoring Reduce Reoffending?

JANUARY 2019

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our time. Our key objective is to build bridges between academic research, policymakers and society.IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

Schaumburg-Lippe-Straße 5–953113 Bonn, Germany

Phone: +49-228-3894-0Email: publications@iza.org www.iza.org

IZA – Institute of Labor Economics

DISCUSSION PAPER SERIES

IZA DP No. 12122

Can Electronic Monitoring Reduce Reoffending?

JANUARY 2019

Jenny WilliamsUniversity of Melbourne and IZA

Don WeatherburnBureau of Crime Statistics and Research

ABSTRACT

IZA DP No. 12122 JANUARY 2019

Can Electronic Monitoring Reduce Reoffending?*

This research evaluates the impact of electronic monitoring as an alternative to prison on

reoffending. Leveraging plausibly exogenous variation in sentencing outcomes generated

by quasi- random assignment of judges, we find electronic monitoring reduces reoffending

within 24 months by 16 percentage points compared to serving a prison sentence. For

offenders who are less than 30, the reduction is 43 percentage points, with sizeable and

significant reductions in reoffending persisting for 8 years. Our calculations suggest that

criminal justice costs are reduced by around $30,000 for each eligible offender who serves

their sentence under electronic monitoring rather than in prison.

JEL Classification: K42

Keywords: electronic monitoring, prison, reoffending, crime

Corresponding author:Jenny WilliamsDepartment of EconomicsThe University of Melbourne3010 VictoriaAustralia

E-mail: jenny.williams@unimelb.edu.au

* The authors thank seminar participants at the University of California (Davis), RAND, University of Melbourne, and

the Bureau of Crime Statistics and Research, David Byrne, Jason Hainsworth, David Marcotte and participants of the

2018 Risky Behaviors Workshop (Corsica) for helpful feedback, and to the teams at the Bureau of Crime Statistics and

Research (especially Suzanne Poynton and Mark Ramsay) and at the Corrections Research Evaluation & Statistics Unit

for providing the data on which this paper is based.

1 Introduction

The incarceration rate has grown substantially in the United States, tripling between 1980 and 2010.

The U.S. is not alone in this regard. The incarceration rate has increased in many OECD countries,

particularly those in western Europe, as well as in countries such as Australia and New Zealand.1

The high cost associated with growing prison populations, estimated to be on the order to 80 billion

dollars in the U.S. (DeVuono-Powell et al. , 2015), has led to substantial interest in sentencing

alternatives to incarcerating offenders in prison. One alternative that has attracted significant

attention is the use of electronic monitoring. Electronic monitoring typically involves offenders

serving their sentence in their home, with their location tracked using an electronic monitoring ankle

bracelet. As such, it provides a low cost means of depriving offenders of their liberty. Nonetheless,

there are concerns that allowing offenders to serve their sentence in their home rather than in

prison is likely to perceived as a “soft” or lesser punishment, and may therefore be less effective at

deterring reoffending.2 We study whether this is the case, providing new evidence on the impact

on reoffending of serving a sentence under electronic monitoring rather than in prison.

Electronic monitoring is not new. Its use in the UK dates back almost thirty years (Hucklesby &

Holdsworth, 2016). In recent times, however, its popularity has grown considerably. In the United

States, the number of accused and convicted criminal offenders subject to electronic monitoring

is estimated to have risen nearly 140 per cent over the 10 year period ending in 2015 (The Pew

Charitable Trusts, 2016). The use of electronic monitoring has also grown throughout Europe,

Australia, New Zealand, Canada and in parts of South America. Despite this, there are few studies

that provide convincing evidence as to whether electronic monitoring reduces recidivism relative to

imprisonment (Renzema & Mayo-Wilson, 2005). This is because those who receive a sentence to

be served under electronic monitoring are likely to differ in observable ways, such as committing

less serious offences, and in unobservable ways, such as exhibiting greater remorse, that make them

less likely to re-offend. As a consequence, the causal effect of serving a sentence under electronic

monitoring rather than in prison on reoffending is difficult to identify.

A second difficulty with research seeking to evaluate the impact of electronic monitoring on

recidivism is that electronic monitoring is not a single well-defined form of intervention applied at

1See Bhuller et al. (2016) for a discussion of the growth in the rate of imprisonment in western Europe.2We focus on the impact of an individual serving a sentence under electronic monitoring rather than in prison

on their own future reoffending, referred to as specific deterrence. We note that the impact of the use of electronicmonitoring rather than imprisonment on deterring crime more broadly, or general deterrence, may also be of concern.While we are unable to address this issue, in their recent review of the literature on general deterrence, Chalfin& McCary (2017) conclude that ”within the range of typical changes to sanctions in contemporary criminal-justicesystems, the evidence suggests that the magnitude of deterrence owing to more severe sentencing is not large”, andthat ”the current elasticity of crime with respect to prison populations is approximately 0.2”.

1

a particular stage in the criminal justice process to a defined set of offenders (DeMichele, 2014).

Electronic monitoring has been used as an alternative to pre-trial detention, as an alternative to

a custodial sentence served in prison, as a condition of bail, as a requirement of community and

suspended sentence orders and as a form of early release. In some jurisdictions it can be used

for any type of offence, including violent offences, whereas other jurisdictions limit its use to non-

violent and less serious offences. In this paper, we study electronic monitoring used as a front-end

alternative to a custodial sentence served in prison for non-violent offenders.

In order to evaluate the impact on reoffending of serving a sentence under electronic monitor-

ing we draw on detailed individual level administrative data from courts, prisons, and corrective

services for a large urban population. An instrumental variables strategy is employed to address

selection into electronic monitoring. Our approach exploits plausibly exogenous variation in sen-

tencing outcomes generated by quasi-random assignment of cases to judges who differ in their

tendency to use the sentencing options of electronic monitoring and prison.3 Our baseline model

examines reoffending within 24 months for offenders of all ages. Estimates from this specification

show that that serving a sentence under electronic monitoring rather than in prison reduces reof-

fending from 58% to 42% (16 percentage points). This effect is robust to accounting for time spent

under electronic monitoring and is not driven by those who have been previously imprisoned. The

reduction in reoffending is, however, found to be concentrated amongst offenders who are less than

30 years old. For this group, serving a sentence under electronic monitoring rather than in prison

reduces reoffending within 24 months from 64% to 21% (43 percentage points).

We also investigate the extent to which the reduction in reoffending that occurs within 24

months of free-time persists or fades out at longer durations of follow-up. To do so, we examine

reoffending out to a ten year horizon. When considering all aged offenders, we find that the

reduction in reoffending persists to five years of free-time. For those who are under 30, the reduction

in reoffending persists to 8 years of free-time.

Finally, we combine our estimates with criminal justice cost information to provide a sense

of the budgetary implications of diverting offenders from prison into electronic monitoring. Our

back-of-the-envelope calculations show that each offender who serves their sentence under electronic

monitoring rather than in prison saves the public purse close to $30,000 on reduced supervision

and future court and prison costs.

Ours is not the first study to evaluate the impact of electronic monitoring on reoffending rela-

tive to prison. Di Tella & Schargrodsky (2013) use a similar identification strategy as this study

3This approach is similar to that used by others to study the impact of incarceration (Aizer & Doyle, 2015; Bhulleret al. , 2016; Dobbie et al. , 2018; Green & Winik, 2010), length of incarceration (Kling, 2006) and pre-trial electronicmonitoring (Di Tella & Schargrodsky, 2013).

2

to examine the impact of pre-trial electronic monitoring in Argentina, Henneguelle et al. (2016)

evaluate a pilot of electronic monitoring as a front-end alternative to prison in France using ge-

ographic and temporal variation in its availability during the role out, and Marie (2009) uses a

regression discontinuity approach to examine the impact of electronic monitoring on reoffending

in the context of early release for offenders serving prison sentences in England and Wales. All of

these studies are unanimous in finding that electronic monitoring reduces reoffending.

In context of this nascent literature, our contributions are three-fold. First, we provide evidence

that an electronic monitoring program that diverts offenders from prison, and that is operating at

scale, reduces reoffending within 24 months of time spent living within the community. Previous

research has studied the use of electronic monitoring at the pre-trial phase, where bail might oth-

erwise be used, and for early release from prison, where parole might be considered an alternative.

Studying the use of electronic monitoring as a front-end alternative to prison is important because,

used in this way, it diverts offenders from prison and thereby serves as a genuine alternative to in-

carceration. Further, the electronic monitoring program we study is well established and operating

at scale. This is important as positive findings from pilots are not always replicated when programs

are scaled up.4

Our second contribution is that we provide the first evidence on the impact of a serving a sen-

tence under electronic monitoring out to a ten year horizon. Previous evidence typically considers

reoffending within 24 months. Our results show that there are significant reductions in reoffending

at much longer durations. This suggests that the diversionary and rehabilitative aspects of the

electronic monitoring program we study led to long term changes in offender behaviour. This is

especially the case for younger offenders, who have potentially longer future offending careers.

Our final contribution is demonstrating that, in addition to the reducing reoffending, the elec-

tronic program we evaluate produced significant criminal justice cost saving for each offender di-

verted from prison. This is important evidence on the potential for the electronic monitoring to

reduce the economic burden or crime.

The remainder of this paper is set out as follows. Section 2 provides background on electronic

monitoring program we study, and the courts and sentencing environment in which electronic

monitoring is available as an alternative to prison. Section 3 presents our conceptual framework

and empirical strategy. Section 4 discusses the data we draw on and provides a descriptive analysis

of the relationship between electronic monitoring, prison and reoffending. Section 5 investigates the

validity of identifying assumptions we employ. Section 6 provides our baseline analysis, a sensitivity

analysis, an examination of a potential violation of the exclusion restriction, an examination of the

4Further, in both the Argentine and French context, electronic monitoring is available and used as alternative toprison for violent offences including rape and murder, which may not be a scenario considered by other jurisdictions.

3

long run impacts of electronic monitoring, and our costing calculations. We conclude in section 7

with a discussion of the implications of our findings.

2 Background

2.1 Electronic Monitoring in NSW

Electronic monitoring has been available in Sydney, the capital of New South Wales (NSW) and

Australia’s most populous city, since 1997.5 It was introduced as a sentencing alternative to prison

for less serious, non-violent offenders. It provides a means of depriving offenders of their liberty

to a degree similar to a minimum security prison, while allowing them to maintain family ties, to

work, and to access support and treatment services. The aims of the electronic monitoring program

we study are to divert less serious and non-violent offenders from prison while maintaining a high

level of supervision, and to reduce the risk of future reoffending through rehabilitation (Auditor-

General’s Report, 2010).

Electronic monitoring involves offenders serving their sentence in the community rather than in

prison but with restricted freedom of movement. While serving a sentence of electronic monitoring,

offenders live at home, either alone or with others (including their family), and are not permitted to

leave their home except when approval is given by their supervising community corrections officer

for activities such as employment, education, intervention programs and medical appointments. All

activities are scheduled on a weekly basis and compliance with the schedule and other requirements

is monitored via an electronic monitoring anklet worn by the offender, through telephone calls,

and regular home and field visits (for out of home activities) from their supervising community

corrections officer (Studerus, 1997). Offenders must also abstain from alcohol and illicit drugs and

are subject to random testing to ensure compliance.

In addition to diverting less serious offenders from the prison system by having them serve their

sentence in the community, electronic monitoring in NSW also aims to reduce the risk of reoffending

though rehabilitation. The aim of rehabilitation is pursued by designing offender-specific programs

to be undertaken while under electronic monitoring.6 For example, offenders whose sentence is

served under electronic monitoring may be expected to be employed. If they are able to work but are

not employed, they could be required to undertake vocational training courses and job search. For

those who are not able to work, developmental activities are typically encouraged (Studerus, 1997).

5The population of Sydney is around 5 million according to the 2016 census (Australian Bureau of Statistics,2016).

6In designing these programs, all aspects of the offender’s lives are taken into consideration, including family issues,parental responsibilities, drug and alcohol issues, health issues and employment responsibilities and opportunities(Auditor-General’s Report, 2010).

4

In addition, specific personal issues, often drug or alcohol related, are addressed through requiring

the offender to undertake counseling or other support services. Rehabilitation programs are also

available to offenders serving a prison term of greater than 6 months; however, undertaking these

programs is optional in prison. Further, offenders serving sentences under electronic monitoring

are supported in their rehabilitation by a high rate of face to face contact with their supervising

officer, who liaises closely with supporting community agencies.7

2.2 Sentencing in NSW

In the NSW court system, the sentencing of guilty offenders is a two-step process whereby the judge

must first decide, based on the seriousness of the offence and the need to protect the community,

whether a custodial sentence is required and if so, the duration of the custodial term. If the judge

determines that a custodial sentence is appropriate, the second step in sentencing is for the judge

to decide the manner in which the sentence should be served. For non-violent offences for which

the judge determines that a custodial sentence of no more than 18 months is appropriate, electronic

monitoring is available as an alternative to prison.

In the event that the judge determines that the offender should serve their sentence under

electronic monitoring, the offender is referred for assessment for suitability. The referral operates

to stay the execution of sentencing pending the completion of the assessment by NSW Community

Corrections. A Community Corrections Officer prepares a report on the offender and if the report is

favourable, the judge may order the offender to serve the sentence under electronic monitoring. In

preparing their report, the Community Corrections Officer first ensure that the offender is eligible.

Eligibility is based on three factors: a sentence length of no more than 18 months; the offences for

which electronic monitoring is being considered are not related to violent of threatening behaviour;

and the offender’s criminal history does not include violent or threatening offences. For eligible

offenders, assessments consider the impact of electronic monitoring on family and/or co-residents

as well as the offender. If the offender is assessed as posing no substantial threat to the family

or other members of the community, can be managed successfully under the terms of electronic

monitoring, and if the the offender and all co-residents consent, the offender is recommended as

suitable (Studerus, 1997).

While electronic monitoring has been available in NSW since 1997, an organisational change

within the NSW Department of Justice shifted the management of the electronic monitoring pro-

gram to the newly established Community Compliance and Monitoring Group (CCMG) in the

second half of 2007. Under the CCMG, assessments became more punitive and started with the

7To accommodate this, supervising officers have smaller case loads and more flexible working conditions.

5

premise that “everyone is unsuitable and must prove themselves to be suitable” (Auditor-General’s

Report, 2010). It required drug and alcohol testing as part of the assessment process and a posi-

tive drug or alcohol test lead to a negative assessment. Further, compliance monitoring included

unannounced home visits and an increase in the use of random drug and alcohol testing. The

approach of the CCMG was not well received by judges. The combination of the more stringent

assessment process and the fall from favour amongst the judiciary lead electronic monitoring to all

but disappear as a sentencing option. For this reason, this study focuses on the period 2000-2007.

Of importance to this study is the way that cases are allocated to judges in the NSW courts

system. When offenders are charged with a crime in NSW, their case is heard in the court in the

jurisdiction where the crime occurred. In urban locations, such as Sydney, courts are typically multi-

court complexes, with each court comprising several court rooms and a different judge presiding in

each courtroom. Offenders are provided with a date, time, and courtroom within a court (or court

complex) for their hearing. There is no scope for offenders to influence the judge that hears their

case. Judges are rotated across courtrooms within a court, across courts within a court complex,

and across court complexes. The process by which judges are assigned to courtrooms does not

depend on the nature of cases being heard, but rather on practical considerations such as the

availability of judges and the case load at courts. On this basis, the Chief Magistrate of NSW

allocates a pool of judges to each multi-court complex, and judges are then allocated to each court

within the court complex by the Supervising Magistrate.8 Given this process, the assignment of

judge to case is effectively random.

3 Research Design

3.1 Conceptual Framework

The primary reason for the use of prison for non-violent offenders over alternate punishments is its

greater potential for specific deterrence. The idea behind specific deterrence is that the criminal

justice sanction imposed is sufficiently costly so as to deter the individual from engaging in crime

in the future. In the case of a prison sentence, the direct cost of punishment is loss of freedom and

autonomy as offenders are removed from the community and serve their sentence in the custody

of the prisons system. Electronic monitoring seeks to restrict freedom to a degree similar to a

minimum security prison with day release, but keeps offenders living within the community. As

8The Listing and Rostering Co-ordinator at the Chief Magistrate’s Office gathers all the information in relation torequired sittings and magistrate commitments and completes the roster for the following week, allocating magistratesto court locations. This roster is generally published by the Thursday for the following week. This information wasprovided to Don Weatherburn in personal correspondence by the Chief Magistrate of NSW.

6

such, serving a sentence in prison exerts a greater specific deterrence effect than serving a sentence

under electronic monitoring. For this reason, and all else being equal, serving a sentence in prison

should lead to a greater reduction in reoffending than electronic monitoring.

That being said, the empirical evidence on the deterrent effect of prison on reoffending is limited

and mixed.9 For example, Aizer & Doyle (2015) find that incarceration as a juvenile increases the

likelihood of incarceration as an adult while Hjalmarsson (2009) finds that a custodial sentence

reduces the recidivism rate of juvenile offenders. In terms of adult offending, Mueller-Smith (2015)

finds that incarceration increases recidivism, Green & Winik (2010) find that it has no effect on

recidivism, while in the context of Norway, Bhuller et al. (2016) find that it reduces reoffending.10

A potential explanation for the inconsistent findings on the impact of prison on reoffending is

that, in addition to its potential deterrent effect, imprisonment may increase reoffending through

other channels. For example, serving a sentence incarcerated in prison places offenders in the sole

company of other offenders, providing them an opportunity to build their criminal networks and

accumulate criminal human capital (Bayer et al. , 2009). In a similar vein, imprisonment is likely

to reduce attachment to work and family, thereby reducing the cost of reoffending (Apel et al. ,

2010; Dobbie et al. , 2018; Lopoo & Western, 2005; Mueller-Smith, 2015). Both the increase in

criminal capital and networks and the reduction in attachment to work and family suggest that

prison may increase reoffending behaviour.

In contrast, under electronic monitoring, strict conditions and close monitoring make oppor-

tunities for interactions with criminal peers less likely. Moreover, those serving a sentence under

electronic monitoring often live with their family, are expected to work and undertake rehabilita-

tive activities (such as drug and alcohol counseling). Collectively, the monitoring, promotion of

attachment to work and family, and addressing drug and alcohol problems, is likely to reduce the

probability of reoffending for those serving a sentence under electronic monitoring compared to an

otherwise similar individual who serves their sentence in prison.

Overall, while specific deterrence suggests that imprisonment may lead to a lower rate of reof-

fending relative to electronic monitoring, reduced attachment to work and family, and the availabil-

ity of criminal peers suggests that prison may increase the risk of reoffending relative to electronic

monitoring. As we cannot disentangle these mechanisms, all of which may come into play, the

9Nagin et al. (2009) comments that “Remarkably little is known about the effects of imprisonment on reoffending.”In their recent and comprehensive review of the literature from economics and criminology, Chalfin & McCary (2017)conclude that the magnitude of deterrent effects of incarceration are likely small.

10Bhuller et al. (2016) note that the Norwegian prison system differs from that in the US in important ways that arelikely to lead to better post-prison outcomes. Specifically, the Norwegian prison system focuses on rehabilitation andpreparing inmates for life after prison, which includes housing prisoners in low security environments and permittingvisits to families. Further, Correctional Services in Norway works with other agencies to ensure that upon release,offenders have a home and a job.

7

question of whether electronic monitoring reduces reoffending relative to prison is an empirical one,

the answer to which this paper seeks to provide.

3.2 Empirical Framework

In order to evaluate the impact of serving a sentence under electronic monitoring rather than

in prison on reoffending, we model reoffending by individual i, reoffi as a function of serving a

sentence under electronic monitoring, EMi (the reference category is serving a prison sentence), a

vector of control variables that always includes fully interacted court and year of court finalization

fixed effects, Xi, and unobserved individual specific characteristics, νi.

reoffi = βEMEMi + β′Xi + νi (1)

There are challenges in obtaining estimates of the causal effect of serving a sentence under

electronic monitoring rather than in prison, βEM on reoffending. First, the manner in which the

offender’s sentence is to be served is not randomly assigned; offenders who serve their sentence

under electronic monitoring are likely to differ from those who serve their sentence in prison in

unobserved ways that are negatively correlated with reoffending, such as the offender’s remorse or

extenuating circumstance contributing to the offender’s behavior. If this is the case, OLS estimates

will be biased towards finding that serving a sentence under electronic monitoring lowers the rate

of reoffending.

A second issue we face is that while it is the judge who decides if an offender is to serve their

sentence under electronic monitoring, in order to do so the judge must first receive a favorable

assessment from Community Corrections.11 Around 70% of those referred for assessment serve a

sentence under electronic monitoring over the period we examine. This suggests a potential for the

assessment process to contribute to a selection effect through positive recommendations to the case

judge, and therefore in those for whom a judge is observed to sentence to electronic monitoring.

We address these issues of endogenous selection into punishment type using an Instrumental

Variables / Two Stage Least Squares (IV/2SLS) approach where the instrumental variable is the

tendency of a quasi-randomly assigned judge to refer eligible cases for assessment for electronic

monitoring. This approach exploits the exogenous source of variation in the likelihood that an

offender receives electronic monitoring that is generated because (a) judges differ in their tendency

to refer offenders for electronic monitoring, and (b) the assignment of judges to cases within a

11Similarly, in the Argentine criminal justice system studied by Di Tella & Schargrodsky (2013), offenders must beassessed for suitability for electronic monitoring before a judge can make an electronic monitoring order, althoughthe authors do not account for this in their study design.

8

court in a given year is effectively random. It is this variation that is used to identify the impact of

electronic monitoring (relative to prison) on reoffending. This approach alleviates the selection into

electronic monitoring that occurs via the assessment process, as well as selection due to unobserved

offender characteristics because the decision to refer to assessment is made by the judge prior to

assessment, and assignment of judge to case is as good as random.12

To be more concrete, we use 2SLS to estimate the causal impact of serving a sentence under

EM rather than in prison, where the first stage is given by:

EMi =αEMπEMij + αXi + εEM

i (2)

and πEMij is the tendency of judge j assigned to offender i to refer eligible cases for assessment for

EM. The structural model for the reoffending is given by (1) above.

2SLS provides consistent estimates of the causal parameters of interest if (1) the instrumental

variable, judge tendency to refer for assessment for electronic monitoring, is relevant in explaining

whether individual i before judge j receives a sentence to be served under electronic monitoring

(relevance), (2) conditional on court by year fixed effects, the judge tendency to refer to assess-

ment for electronic monitoring is as good as randomly assigned (random assignment) , and (3) the

judge tendency to refer for electronic monitoring assessment only impacts on reoffending through

its impact on receiving a sentence of electronic monitoring (excludability). Further, under het-

erogeneous treatment effects, a Local Average Treatment Effects (LATE) interpretation requires

(4) that the instrument satisfies monotonicity. In our application, this means that the probability

that individual i receives as sentence of electronic monitoring is monotonically increasing in the

tendency of judges to refer eligible cases for assessment for electronic monitoring. We examine the

conditions of relevance, conditional independence and monotonicity in Section 4. We come back

to the exclusion restriction later in the paper. If these assumptions are met, our empirical design

identifies the causal impact of serving a sentence under electronic monitoring for individuals who

would have served their sentence in prison had they faced a different judge. This is the Local

Average Treatment Effect (LATE).

In our baseline model, the window over which reoffending is measured is 24 months of free-time,

where free-time is defined as time spent in the community. For those who serve their sentence in

prison, free-time begins at the date of release. For those who served their sentence under electronic

monitoring, free-time begins at the date that their case is finalized because a sentence of electronic

12Differences across courts in the utilization of electronic monitoring has been noted repeatedly since the early daysof its introduction. See for example (Heggie, 1999). While there may be a variety of reasons why judges (and hencecourts) differ in their willingness to use electronic monitoring rather than imprisonment, the underlying reasons donot matter; the validity of the approach rests on the random assignment of judges to cases (Bhuller et al. , 2016).

9

monitoring is served in the community. While living in the community permits reoffending for

both prison releasees and those who are serving their sentence under electronic monitoring, it is

important to recognise that the later group are subject to a greater degree of specific deterrence

while serving their sentence. This is because (1) detection of reoffending is highly likely while under

the monitoring; and (2) reoffending is a serious breach of the conditions of electronic monitoring

and, irrespective of the type of offence, results in immediate referral to the Parole Board and likely

revocation of electronic monitoring, with a prison term equal to the balance of the original prison

sentence less time served under electronic monitoring imposed.13 This swift and certain punishment

creates a specific deterrence effect for those serving a sentence under electronic monitoring for the

duration of the monitoring sentence that is absent for prison releasees.14 In order to get a sense of

the importance of this on our baseline findings, we consider an alternative measure of reoffending

in which free-time begins at the date of completion of electronic monitoring sentences, and at

the date of release for prison sentences. Similarly, the specific deterrent effect of prison may be

diminished by previous experience of imprisonment, including while awaiting trial (Dobbie et al. ,

2018).15 Alternatively, those who have been previously incarcerated may view serving a sentence

under electronic monitoring as a “second chance”, to which they respond by reducing reoffending

(Henneguelle et al. , 2016). We investigate the extent to which our findings are impacted by previous

incarceration in our sensitivity analysis.

4 Data

The data used in this analysis is drawn from two sources. The first is the Reoffending Database

(ROD) provided by the NSW Bureau of Crime Statistics and Research (BOCSAR). ROD links

individual case level data on all matters before Local, District and Supreme Criminal Courts,

and Childrens Criminal Courts in New South Wales (NSW), with adult and juvenile custody data.

Information on the judge overseeing each case was linked to the ROD data by BOCSAR. The second

source of data used in this study is information on referrals of offenders to the NSW Community

Corrections for assessment for suitability for electronic monitoring. These data was provided by

the Corrections Research Evaluation & Statistics Unit (Corrective Services NSW ) to BOCSAR,

13Serious or recurring minor breaches of conditions of electronic monitoring are immediately referred to the ParoleBoard for revocation. There are strict time lines for breach reports to be provided to the Parole Board and for theParole Board to make their determination. In determining whether a breach has occurred, the Parole Board appliesthe administrative standard of proof, which is based on the “balance of probabilities” (Studerus, 1997).

14Evidence on the deterrence effect of swift and certain punishment in the context of Hawaii’s Opportunity Pro-bation with Enforcement (HOPE) program is reported by Hawken & Kleiman (2009) and in the context of SouthDakota’s 24/7 Sobriety Project by Kilmer et al. (2013).

15For example, Dobbie et al. (2018) find that pre-trial detention increases the probability of rearrest following casefinalization, but it decreases it prior to finalization.

10

which linked it to the offender data extracted from the ROD for this study. The custody data are

available from 2000. Given the changes in the administration of the electronic monitoring program

than occurred in the second half of 2007, we focus our analysis on sentencing that occurred over the

period 2000-2007. The follow-up period over which we measure reoffending extends to September

2016.

From the universe of the ROD, we extract all individuals whose primary offence is eligible for

electronic monitoring and who are sentenced to prison or electronic monitoring. This includes

proven offences for which the case judge deems imprisonment of a term of no more than 18 months

is the appropriate punishment, where the sentence may be served in prison or under electronic

monitoring. The requirement of eligibility for electronic monitoring excludes the following offences

from our sample: murder; attempted murder; manslaughter; serious assault resulting in injury; all

sexual assault and related offences; abduction; harassment and other offences against the person;

importing, exporting, dealing or trafficking illicit drugs; and prohibited and regulated weapons and

explosives offences. We further restrict our sample to include only those judges who presided over

10 or more electronic monitoring eligible cases over the period 2000-2007. This ensures that the

judge tendency measure does not include judges who only appear once or twice for relevant cases

over the sample period. As randomization of case to judge occurs at the court level, we also limit

our sample to cases heard in courts for which more than one judge presided in each calendar year.

As electronic monitoring was not generally available outside Sydney (the capital city of NSW)

over the period of our analysis, we also limit observations to offenders living Sydney at the time

they offended. Our final restriction is to limit the sample to offenders who are not identified as

Aboriginal, as Aboriginals have a distinctly different relationship with the criminal justice system.

After applying the above set of restrictions, the sample consists of 16,475 cases. We refer to

this sample as our full sample. Our judge tendency measure is based on the full sample. The

sample used to estimate the structural parameter of interest consists of the first instance in which

an offender in the full sample is before the court for an electronic monitoring eligible offence. This

offence is referred to as the index case, and we refer to the sample of index cases as our estimation

sample. There are 8,826 index cases in our estimation sample. Note that we have court data until

September 2016, so we are able to measure reoffending in our estimation sample out to this horizon.

4.1 Reoffending

The primary outcome of interest is reoffending within 24 months of free-time (time living in the

community). We construct the indicator for reoffending within 24 months of free-time using the

elapsed time between either being released from prison and the first subsequent case before the

11

courts, or in the case of electronic monitoring, between the date of finalization of the index offence

and the first subsequent case before the courts.16 In addition to reoffending, the period of time

an offender spends living in the community may end because he dies, or he is incarcerated for

prior offences. If the offender’s episode of free-time ends due to either of these reasons, or the

follow-up period ends before 24 months of free-time elapses (that is their duration of free time is

censored), the indicator for reoffending within 24 months is coded as missing since we do not know

whether the offender would have reoffended with 24 months. This reduces the number of index

cases in our estimation sample from 8,826 to 7,366.17 Note that we use the term case and offender

interchangeably as there is only a single index case for each offender in our estimation sample.

In an extended analysis we also consider the impact of convicted offenders serving a sentence

under electronic monitoring relative to serving a prison sentence on recidivism within 12 months

through to 120 months. This gives some sense as to whether any difference in reoffending by

those serving a sentence under electronic monitoring is enduring or short lived. Each measure of

recidivism within the given duration of time living in the community is constructed as an indicator

equal to one if the individual reoffends within the period and zero otherwise, with those who die,

are incarcerated for earlier offences, or whose follow-up period is shorter than the duration under

study coded as missing.

4.2 Punishment

We limit our analysis to cases that are deemed punishable by imprisonment and are eligible for

electronic monitoring. Table 1 shows the distribution of crimes in the full sample and the estimation

sample (which consists of the index cases) by the manner in which their sentence is to be served.

There are three main points to take away from Table 1. First, for the full sample we observe

offenders receiving sentences of electronic monitoring and prison for each type of offence that

is eligible for electronic monitoring. Second, for the full sample and the estimation sample the

top three types of offences that are punished with electronic monitoring are traffic, fraud and

government procedure, while the top three for prison are theft, traffic and government procedures.

The third point to take away from Table 1 is that the distribution of offence type for each form

of punishment is similar across the full sample and the estimation sample. For example, in the

full sample 47% of cases punished with electronic monitoring have as their most serious offence

a traffic offence compared to 45% in the estimation sample. Similarly, in the full sample, 20% of

16We include time serving a sentence under electronic monitoring as free-time as the individual is living in thecommunity and is able to reoffend (more readily than if they were in prison).

1785 are censored due to death; 1,355 are censored due to being incarcerated for prior offences; and 20 due to thefollow-up period ending in less than 24 months of free-time.

12

cases receiving a prison sentence are for traffic offences compared to 21% in the estimation sample.

4.3 Control Variables

As noted earlier, all models we estimate control for a full set of interacted court and year of

finalization fixed effects. This accounts for any differences in the types of cases heard, or culture of

strictness, across courts and time, given that randomization occurs within courts at each point in

time. We also control for crime type (injury, negligent driving, break and enter, theft, fraud, public

order, and traffic with violations of government procedures as the omitted category) gender (with

an indicator equal to one if the offender is male) and a quadratic in age. We additionally estimate

models that control for an extended set of information, including the number of finalised court

appearances within five years prior to the current court appearance, an indicator for the offender

having no legal representation during their hearing, and a set of indicators for the local government

area in which the offender lives. Descriptive statistics (sample mean and standard deviation) for

the control variables can be found in Table 4.

4.4 Descriptive Analysis

Table 2 shows the reoffence rate by the manner in which the sentence is served for the estimation

sample at 12 months through to 120 months of free-time. It shows that 47% of offenders in the

estimation sample who are imprisoned face new charges before the courts within 12 months of being

released, and this increases to 84% within 120 months of release. The reoffending rate for those who

serve a sentence under electronic monitoring is 16% within 12 months, increasing to 66% within 120

months of free-time. These descriptive statistics suggests that imprisonment is associated with a

greater likelihood of reoffending at any duration of free-time compared to electronic monitoring. Of

course these associations could simply reflect favourable selection (in terms of risk of reoffending)

into electronic monitoring. In order to determine the causal relationship between sentence type

and reoffending, we follow the instrumental variables approach described in the previous section.

We next turn to describing our instruments.

5 Instruments

5.1 Judge referrals to electronic monitoring

We use the full sample of 16,475 cases satisfying the restrictions discussed above to construct our

measure of judge tendency to refer offenders for assessment for suitability for electronic monitoring.

The measure we use is based on the judge specific leave-out mean rate of referral (Aizer & Doyle,

13

2015). This is constructed as the judge specific average rate of referral for all other cases the

judge has handled, including past and future cases, and not only the index offences that comprise

our estimation sample. The leave-out mean represents the case level probability that the assigned

judge refers the offender to assessment for electronic monitoring based on the judge’s past and future

decisions for all other electronic monitoring eligible cases, excluding the case in question. Following

Bhuller et al. (2016) and Dobbie et al. (2018), we construct the judge tendency measure for

electronic monitoring as the residual from regressing the leave-out mean judge electronic monitoring

referral rate on a set of fully interacted court and year of finalization fixed effects. This controls for

differences across time and across courts in the types of cases heard, or culture of strictness. We

do this as randomization occurs within the pool of available judges at the court level.

There are 139 judges and 52 courts in our estimation sample. The distribution of judge tendency

to refer to assessment for electronic monitoring for the estimation sample is shown in Figure 1. For

ease of interpretation, we re-center the distributions by adding the average of the leave out means to

the residuals. The resultant mean tendency to refer to assessment for electronic monitoring is 0.12

with a standard deviation of 0.08. Figure 1 reveals a substantial range in the tendency of judges

to refer to assessment for electronic monitoring, with judges at the 95th percentile referring to

electronic monitoring 25% of the time compared to 2% of the time for judges at the 5th percentile.

5.2 Instrument Relevance

Table 3 contains the first stage estimates. This table reports the key results from OLS regressions in

which the dependent variable is an indicator equal to 1 if offender i before judge j receives a sentence

to be served under electronic monitoring. The explanatory variable of interest is the instrument;

judge tendency to refer to electronic monitoring assessment. We report several specifications for

this first stage model. In the first, reported in column 1, we control for the set of fully interacted

court and year of finalization indicators only. In the second specification, we add controls for

demographics (an indicator for gender is male, a quadratic in age) and indicators for type of crime

(violation of government procedures is the reference category). The final specification adds further

controls for the number of finalised court appearances within five years prior to the current case,

an indicator for no legal representation in the current case, and indicators for the local government

area (LGA) in which the offender resides. For each specification, we report the coefficient estimates

for the instrument, judge tendency to refer to assessment for electronic monitoring, the results

from a diagnostic test that examines whether the parameters in the model are under-identified,

along with the commonly reported F-statistic testing the significance of the instrument. Note that

standard errors (and test statistics) account for clustering at the court level.

14

As can be seen from Table 3, judge tendency to refer for assessment for electronic monitoring

is a highly significant predictor of the offender receiving a sentence of electronic monitoring. The

point estimates imply that a one standard deviation (0.08) increase in the tendency to refer to

electronic monitoring by the judge faced increases the likelihood that the offender is sentenced to

electronic monitoring by around five percentage points, or 39%.

Table 3 also reports diagnostic statistics on the significance of the instrument, along with the

Sandersom-Winjmeijer under-identification test. As can be seen from Table 3, the standard F-

statistic testing the significance of the instrument in the linear probability model for receiving a

sentence of electronic monitoring are much larger than the rule of thumb value of 10, and this is the

case across all specifications.18 Also, the Sanderson-Windmeijer test leads to the rejection of the

null hypothesis that the model is under-identified in all specifications. From Table 3 we conclude

that the model is not under-identified, and that the instrumental variable is relevant, and not weak.

5.3 Instrument Validity

Key to the validity of our instrumental variable approach is the assumption that the instrument is

as good as randomly assigned and therefore uncorrelated with unobserved offender characteristics

that are related to the likelihood of reoffending. While we cannot directly test the assumption that

judge tendency to refer for assessment for electronic monitoring is randomly assigned (conditional

on the interacted court and year fixed effects), the results in Table 3 provides some evidence on

this issue. If judges are as good as randomly assigned, then adding the pre-determined variables

(that are used as controls) should not significantly change the first stage coefficient estimates on

the instrument. As can be seen by looking across the specifications, as controls are added to the

first stage regression model, the point estimates of the coefficient on the judge tendency to refer to

electronic monitoring assessment do not change appreciably.

A further implication of random assignment of judge to case within courts, is that judge tendency

to refer to electronic monitoring assessment should not be correlated with observable characteristics

of offenders or their case (conditional on the court and year interacted fixed effects). We provide

evidence on this in Table 4. Specifically, Table 4 shows that while receiving a sentence of electronic

monitoring is correlated with characteristics of the offender and case conditional on interacted court

and year of finalization fixed effects(F-stat=31.29, p-value=0.000), judge tendency to refer cases

for electronic monitoring assessment is not (F-stat=1.40, p-value=0.20).

18As the outcome of interest, reoffending, does not have iid errors, we are unable to test whether the instrumentalvariables are weak using Stock-Yogo values. This is because there are no tabulated values for the sampling distributionwith which to compare the test statistic for weak instruments when errors are not iid (see Windmeijer for more onthis)

15

While conditional random assignment of the instruments is sufficient for the reduced form

estimates to have a causal interpretation, the exclusion restriction is additionally required in order

for the IV estimator to provide causal estimates of the structural parameter of interest. We present

an investigation of this issue following the discussion of our baseline findings.

5.4 Monotonicity

If the impact of receiving a sentence of electronic monitoring on reoffending is heterogenous, a

LATE interpretation of the IV estimates requires monotonicity. In our application, monotonicity

implies that an offender who receives a sentence of electronic monitoring from a judge who has a

low tendency for referral will also receive a sentence of electronic monitoring from a judge with

a high tendency for referral. If this condition is satisfied, then our estimates uncover the impact

of serving a sentence under electronic monitoring among the sub-group who would have received

incarceration if their case was heard by a different judge with a lower tendency to refer to electronic

monitoring assessment.

While we cannot directly test whether monotonicity holds in our sample, we follow previous

studies, and test two implications of monotnicity (Aizer & Doyle, 2015; Bhuller et al. , 2016; Dobbie

et al. , 2018). First, the impact of judge tendency to refer to assessment for electronic monitoring

on the probability that an offender receives a sentence to be served under electronic monitoring

should be non-negative for any sub-sample. To examine whether this is the case, we construct the

judge tendency variable using the full sample, but estimate the first stage on sub-samples defined

by crime types, by whether the offender has been incarcerated in the past 5 years, and by age. The

results are reported in Table 5. They show that for all sub-samples, judge tendency for referral is

positive and statistically significant in the first stage model for the offender serving their sentence

under electronic monitoring.

The second implication of monotonicity that we investigate is that judges who are more inclined

than others to refer to electronic monitoring assessment for one crime type should also be more

inclined to do so for other crime types. For example, judges who are more likely to refer traffic

offences to electronic monitoring assessment should also be more likely to refer other crimes types

to electronic monitoring assessment. If this is true, then the probability that an offender facing

a non-traffic charge receives a sentence to be served under electronic monitoring should be non-

negatively related to the (same) judge’s tendency to refer traffic offences to electronic monitoring

assessment, referred to as the ‘reverse instrument’.19

19Bhuller et al. (2016) refers to this method of investigating the monotonicity assumption as reverse sampleinstruments since it involves instrumenting the endogenous variable for a sub-sample using an instrument constructedfrom the complement of the sub-sample.

16

The second panel of Table 5 reports the first stage results based on the reverse instruments

defined by crime type, previous incarceration and age. In all cases, the reverse instruments for

referral to assessment for electronic monitoring are significantly positively related to the probability

of receiving a sentence of electronic monitoring. Moreover, the coefficient estimates are of a similar

magnitude to those for the baseline instruments. Overall, we conclude that there is no evidence to

suggest that the assumption of monotonicity is rejected in these data.

6 Effect of Electronic Monitoring on Recidivism

6.1 Baseline Results

Table 6 presents 2SLS estimates of the impact of serving a sentence under electronic monitoring

relative to prison on the probability of reoffending within 24 months of free-time. The table also

reports OLS estimates as a point of comparison, as well as reduced form estimates. The first column

of Table 6 contains OLS, reduced form and 2SLS coefficient estimates based on a specification that

controls for court and years of finalization interacted fixed effects only. The second column adds

demographic characteristics of the offender, and crime type to the list of controls. The third

column reports on our preferred specification, which additionally controls for the number of court

appearances in the five years prior, having no legal representation during the court hearing, and

indictors for the local government area of offender residence. Standard errors are clustered at the

court level to account for the fact that quasi-randomisation with respect to judge occurs at the

court level.

Beginning with the OLS estimates, we find a significant negative association between the proba-

bility of reoffending within 24 months of free-time and receiving a sentence of electronic monitoring

relative to receiving a prison term in all specifications. When only court and year of finalization

interacted fixed effects are accounted for, the point estimates suggest that serving a sentence under

electronic monitoring is associated with a 25 percentage point reduction in the likelihood of reof-

fending compared to serving it in prison. This is a sizeable reduction given that the average rate

of reoffending within 24 months of free-time amongst those who serve a prison sentence is 58%.

Looking across the row, we see that accounting for the offenders demographic characteristics and

the type of crime the offender is charged with significantly reduces the OLS point estimates, and

additionally accounting for the number of finalized court appearances in the previous five years, a

lack of legal representation during the court case and the local government area of their residence,

reduces the OLS point estimates even further. Accounting for the full set of controls reduces the

magnitude of the OLS coefficient on electronic monitoring by 40% relative to its magnitude in the

17

specification that only controls for court and year interacted fixed effects. This suggests that judges

take these observable factors into account in choosing the type of sentence they hand down. In

other words, judges are choosing the alternative of electronic monitoring for those who are less

likely to reoffend on the basis of their observed characteristics.

Similarly, the reduced form estimates show that accounting for observable characteristics leads

to sizeable reductions in the estimated impact of judge tendency for referral for electronic monitoring

assessment on reoffending. For example, the impact of an increase in judges tendency to refer to

assessment from 0 to 1 is estimated to reduce reoffending (within 24 months of free-time) by 17

percentage points in the specification controlling for court and year interacted fixed effects only,

compared to 9 percentage points in the full specification that additionally controls for demographics,

crime type, criminal history, legal representation and local government area of residence. The

reduced form estimates of the impact of judge tendency for referral for assessment for electronic

monitoring remains statistically significant at the 1% level even after controlling for the full set of

covariates.

The IV estimates are reported in the bottom section of Table 6. These results provide robust

evidence that serving a sentence under electronic monitoring reduces the probability of recidivism

within 24 months of free-time between 27 percentage points (when only court and year of finalization

interacted fixed effects are controlled for) and 16 percentage points (when the full set of controls

are included). Given that the average rate of reoffending (within 24 months of free-time) amongst

those who were incarcerated is 58%, the reduction in reoffending attributable to serving a sentence

under electronic monitoring compared to imprisonment is between 47% in the specification that

controls for court by year fixed effects only and 28% in our preferred specification (which includes

the full set of controls).

6.2 The Exclusion Restriction

In order for our IV estimates to have a causal interpretation, the judge tendency to refer for

electronic monitoring assessment must only impact on reoffending through the channel of receiving

the punishment of electronic monitoring. A potential concern regarding this assumption is that

judges may trade off severity of sentence and sentence length, or alternately, that judges who are

more severe in their choice of punishment type may also be more severe in terms of punishment

length (Aizer & Doyle, 2015).

These scenarios raise two potential threats to the exclusion restriction. The first occurs if judge

tendency for sentence length has a direct effect on reoffending and is correlated with judge tendency

to refer for assessment for electronic monitoring. If this is the case then the exclusion restriction

18

is violated because the instrumental variable, judge tendency to refer for electronic monitoring

assessment, is correlated with the omitted (from the structural equation) regressor, judge tendency

for sentence length. This source of bias can be addressed by including judge tendency for sentence

length in the set of control variables.

A second way in which the baseline model may produce unreliable estimates of the causal

parameter of interest is if the sentence length received directly impacts on reoffending (conditional

on the manner in which the sentence is served and the control variables), and the sentence length

received depends on the judge tendency to refer offenders for electronic monitoring assessment

(conditional on judge tendency for length of sentence). The exclusion restriction is violated in

this case because the instrument (judge tendency to refer for assessment for electronic monitoring)

impacts on reoffending through (omitted) sentence length as well as sentence type. This potential

source of bias can be addressed by augmenting the baseline model to include sentence length as

an endogenous regressor and instrumenting it with judge tendency for sentence length (defined as

the mean adjusted residual from regressing the leave-out mean sentence length on a set of fully

interacted court and year fixed effects) as follows:

EMi =αEMπEMij + αlgπ

lgij + αXi + εEM

i

LENGTHi =δEMπEMij + δlgπ

lgij + δXi + εlgi

reoffi =βEMEMi + βlgLENGTHi + β′Xi + νi

where LENGTHi is the sentence length received by offender i and πlgij is the sentence length

tendency of judge j assigned to offender i. Because cases are randomly assigned to judges, judge

tendency for sentence length in addition to their tendency to refer to assessment for electronic

monitoring satisfy the independence assumption.

Table 7 shows the results from including the judge tendency for sentence length as an additional

control variable in the baseline model (column 2) and by augmenting the baseline model to include

sentence length as an endogenous regressor and instrumenting it with judge tendency for sentence

length (column 3). We also repeat the results for the baseline model for ease of comparison (column

1). As is clear from the results reported in the table, there is no evidence to suggest that the

exclusion restriction is violated due to judge tendency for sentence length impacting reoffending

as shown in Column 2. In particular, the judge tendency for sentence length is uncorrelated

with reoffending (conditional on the manner in which the sentence is served and the other control

variables), and accounting for judge tendency for sentence length as an additional control has little

19

impact on the IV estimates of serving a sentence of electronic monitoring, with the point estimate

increasing slightly in magnitude from -0.16 to -0.17. Similarly, when accounting for sentence length

as an endogenous determinant of reoffending, judge tendency for electronic monitoring does not

predict sentence length, and the coefficient estimate of the impact of sentence length on reoffending

is not statistically significant. Overall, we conclude that it is unlikely that the exclusion restriction

is violated through judge tendency for punishment type impacting on reoffending through sentence

length.

6.2.1 Robustness

As previously discussed, there may be some concern that including time spent serving a sentence

under electronic monitoring as free-time may lead to an overly optimistic estimate of the impact of

serving a sentence under electronic monitoring. In order investigate this issue, we measure free-time

for those serving their sentence under electronic monitoring from the time their sentence ends.20

As can be seen from the results in the first column of Table 8, measuring free-time for all offenders

from the time their sentence ends increases the magnitude of the estimated impact of serving a

sentence under electronic monitoring on the probability of reoffending from -0.16 to -0.19.

Previous research has found that the reduction in reoffending from the use of electronic moni-

toring differs for those who have previously served a prison sentence compared to those who have

not (Henneguelle et al. , 2016). In order to investigate whether this is the case in the context

we study, we re-estimate our baseline model over the sub-sample of offenders who have not been

imprisoned in the 5 years prior to the index offence. The results from doing so are reported in

column 2 of Table 8. As can be seen from the table, the magnitude of the estimated impact on

reoffending of serving a sentence under electronic monitoring increases slightly from -0.16 to -0.20

when we remove those who have served a prison sentence (in the last 5 years) from the sample.

Similarly, the impact of serving a sentence under electronic monitoring may differ for those

who have been detailed pre-trial and those who have are remain living within the community. To

examine if this is the case in our sample, we remove observations from the sample on individuals

who were detained while awaiting trial for their current case in addition to those who had been

imprisoned in the previous 5 years. The results from doing so are reported in column 3. They

show that receiving a sentence of electronic monitoring for those who have not been in prison and

were not detained pre-trial reduced the likelihood of reoffending within 24 months of free-time by

20We note that this not an ideal solution as 73 of those serving an electronic monitoring sentence (7.3% of the EMsample) reoffend while serving their sentence, with half of receiving a prison sentence for the new offense in additionto serving the balance of their original sentence in prison. For these 36 individuals, we do not observe free-timefollowing their original sentence, and so they are dropped from the analysis. For the 37 individuals who do notreceive a subsequent prison term, we measure free-time from the end of the sentence for the index offense.

20

13 percent, similar to our baseline estimated reduction of 16 percent.

Finally, we examine the extent to which the impact on recidivism of serving a sentence under

electronic monitoring differs for younger and older offenders. To do so, we split the estimation

sample into cases for which the offender is less than the median age of 30, and those for which

the offender is at least 30 years old, and estimate over these subsamples. The results from doing

so are reported in columns 4 (less than 30) and 5 (age is at least 30) of Table 8. They show that

the reduction in reoffending is solely attributable to those offenders who serve their sentence under

electronic monitoring who are less than 30 years old. This groups reoffending is reduced by 43

percentage points, which is significantly larger than the estimated impact for all ages. In contrast,

there is no difference in reoffending for those who serve their sentence in prison or under electronic

monitoring amongst those aged 30 or older.

From these analyses we conclude that the baseline results regarding the impact of electronic

monitoring on reoffending are not driven by including in our measure of free-time, time in the

community serving a sentence under electronic monitoring. In addition, we find no evidence that

the impact of electronic monitoring on reoffending is greater for those who have previously been

imprisoned, including those whose incarceration occurred while awaiting finalization of their case.

We do, however, find significant evidence that the impact of electronic monitoring on reoffending

is driven by offenders aged under 30.

6.3 Long Run Impacts

Table 9 explores whether the reduction in reoffending from serving a sentence under electronic

monitoring rather than in prison endures through to a time horizon of ten years of free-time, or

whether it fades out. We also investigate whether the differential effects for younger (aged less

than 30) and older (aged at least 30) offenders are robust to the time horizon considered. The

table reports on specifications for which the outcomes of interest are reoffending at 12 monthly

intervals ranging from 12 months to 120 months of free-time, and that account for the full set of

control variables. Panel A contains estimates based on the full estimation sample, Panel B reports

estimates based on the subample of offenders who are aged less than 30 at their index offense, and

Panel C reports estimates based the subample of offenders who are aged at least 30 at their index

offense. Note that there is only a very modest amount of attrition at the longer follow-up period

of 120 months of free-time relative to our baseline results of 24 months of free-time.21

As can be seen from Panel A of Table 9, using the full sample, serving a sentence under electronic

21The attrition rates are 8% for the full sample, 6% for the less than 30 subsample, and 11% for the 30 and oldersubsample.

21

monitoring rather than in prison is estimated to reduce reoffending by 20 percent points after 12

months of free-time and by 15 percentage points after 60 months, and these effects are statistically

significant at conventional levels. The reduction in reoffending is estimated to be be 12%, 11%

and 10% at 72, 84 and 96 months respectively, falling to 3% at 120 months of free-time, although

these estimates drop below conventional levels of significance. The estimates in Panel B show that,

for offenders who aged less than 30 at the time of their index offense, the reduction in reoffending

is larger than in the full sample, estimated to be 41% after 12 months of free-time, 35% after 60

months, and 31% after 96 months of free-time, with statistically significant effects at 12 , 24, 84 and

96 months. For this younger group, the impact of serving a sentence under electronic monitoring

falls to 22% at 108 and 120 months of free-time, although these estimates are not significant at

conventional levels. The estimates in Panel C show that amongst those who are 30 years of age

or older, there is no significant difference in reoffending amongst those whose serve their sentence

under electronic monitoring and those who serve it in prison at all durations of free-time from 12

months to 120 months.

Overall, these estimates show that the impact of serving a sentence under electronic monitoring

rather than in prison remains large and statistically significant, even at five years of free-time based

on the full sample of offenders, and at eight years amongst offenders aged less than 30 at the time

of the index offense.

6.4 Intensity, Seriousness, and Punishment for Reoffending

Our results provide robust evidence that serving a sentence under electronic monitoring reduces the

likelihood of reoffending amongst eligible offenders who, had they faced a judge less inclined to use

electronic monitoring, would have received a prison sentence. All else being equal, this provides a

strong evidence base that policy makers can draw on to modify current sentencing policy for non-

violent, and less serious offences. However, there may be more than one dimension along which

serving a sentence served under electronic monitoring impacts on future offending behaviour that is

relevant for informing policy makers. For example, serving a sentence under electronic monitoring

may impact on the number of subsequent crimes committed, the seriousness of subsequent crimes,

and the likelihood of future incarceration.

Our data provide some opportunity for examining the impact of serving a sentence under

electronic monitoring on these more nuanced outcomes. We begin by considering the impact of

receiving electronic monitoring on the intensity of reoffending, as measured by the number of times

the offender is before the courts in the 24 month period of free-time following the index case. Given

the potential for selection into reoffending we define our dependent variable for this analysis as

22

either a count of the number of times that the offender is before the courts, including zero for those

who do not re-offend, or the log of the number of times the offender is before the courts.22

The results are shown in the last two columns of Table 8. When the outcome variable is the log

of the number of court appearances, we estimate that receiving a sentence served under electronic

monitoring reduces the number of subsequent court appearances by 50%. Evaluated at the sample

mean of 2.5 court appearances, this translates into a reduction of 1.2 subsequent court appearances.

When the outcome is the number of court appearances, we estimate that serving a sentence under

electronic monitoring rather than in prison reduces the number of subsequent court appearances

by around 1.1, which is in agreement with the estimate from the model in which the outcome is

measured in log form.

We would also like to be able to determine if serving a sentence under electronic monitoring

impacts on the seriousness of subsequent offences, and therefore the burden of subsequent reoffend-

ing on the criminal justice system, as measured by the degree of supervision required. Table 10

shows the empirical distribution of reoffence type (no reoffence, less serious reoffence, and serious

reoffence) by whether the index offence attracted a sentence to be served under electronic moni-

toring or in prison.23 As can be seen from the table, only 8% of those who serve their sentence

under electronic monitoring reoffend with a serious offence within 24 months compared to 17% for

those who served a prison sentence. Combined with the fact that only a small fraction of eligible

offences in our estimation sample receive electronic monitoring, we conclude that these data do

not provide sufficient variation to separately identify the impact of electronic monitoring on serious

reoffending relative to not reoffending or less-serious reoffending.24 It is interesting to note from

Table 10 that, conditional on reoffending, the probability of committing a serious offence is similar

for those receiving a sentence served in prison (29%) or under electronic monitoring (27%).

Given the infrequent use of electronic monitoring by judges, and the low rate of observed reof-

fending amongst those who serve their sentence under electronic monitoring, there is also insufficient

variation in the data to identify the impact of electronic monitoring on subsequent sentencing. Ta-

ble 11 shows that amongst those who served their sentence under electronic monitoring, 10% were

subsequently sentenced to a custodial sentence within 24 months, compared to 25% of those who

served a prison sentence for the index offence. Table 11 also shows that, once we condition on

reoffending, the proportion serving a custodial sentence for a subsequent offence is 35% for those

22In order to define the logged outcome variable over all offenders, rather than the selected sample of re-offenders,we re-center the number of court appearances by adding 0.1 before taking the natural log.

23Serious offences are defined as homicide, injury, sexual assault, negligent acts, abduction and robbery. Lessserious offences are defined as break and enter, theft, fraud, drugs, weapons, property damage, public order, traffic,government procedure and miscellaneous offences.

24Amongst the 996 individuals who served a sentence under electronic monitoring for the index offence, 294 sub-sequently committed an offence, and just 39 committed a serious offence within 24 months.

23

whose index sentence is electronic monitoring compared to 44% for those whose index sentence is

prison.

All in all, the evidence points to electronic monitoring impacting on whether or not a person

reoffends, and the intensity of their reoffending, but perhaps not on the seriousness of offending

amongst those who do reoffend, and this is reflected in similar proportions serving a custodial

sentence amongst reoffenders.

6.5 Cost Savings

Based on these findings, and using costing information on electronic monitoring and prison provided

in the 2010 Auditors-General’s Performance Audit Report, and the 2010 Report on Government

Services, we are able to conduct some back-of-the-envelope calculations that illustrate the order of

magnitude of the cost savings from using electronic monitoring rather than prison in sentencing

eligible offenders. As shown in Table 12, we focus on savings per offender who serves a sentence

under electronic monitoring attributable to the corrective services cost differential for electronic

monitoring versus medium security prison, and those arising from reduced future court appearances

and subsequent sentences served under incarceration.25

Table 12 shows the largest part of the savings per offender is from the direct cost of the index

offence sentence. These savings are estimated to be of $25,200 per offender diverted from prison

by serving a sentence under electronic monitoring. This figure assumes the average sentence length

of 180 days (6 months) in our data and uses a per day cost of $47 for those serving a sentence

under electronic monitoring and $182 for an individual serving a sentence in a medium security

prison, as used by Auditor-General’s Report (2010). Additional savings come from one fewer court

appearances and reduced subsequent incarceration in the two year follow-up period.26 We calculate

the reduced cost due to averted subsequent custodial sentences of incarceration by assuming a

reduction in the probability on reoffending of 0.16 (from our baseline estimates), and that the

probability of a custodial sentence conditional on reoffending is 0.44 (which is the probability in

the sample who are imprisoned for the index offence given in Table 11). We use the average

sentence length for custodial sentences in the reoffending sample of 260 days, and a per day cost

of incarceration of $217 (which is the average for secure prisons).

Collectively, these figures produce a cost saving per person who serves their sentence under

electronic monitoring rather than in prison of $29,252. If we apply this cost saving to the annual

25The cost per case in the Magistrates’ courts in 2008-2009 is $495 (table 7A.23) and the net operating expenditureper prisoner per day in a secure prison in 2008-2009 is $217 per day (table 8A.7) Productivity Commission (2010).

26In calculating these savings we assume the additional court hearing and subsequent custodial sentence occur inthe second year and these costs discounted back to the initial period (assuming a discount rate of 5%).

24

average number under of offenders under electronic monitoring over the period 2000-2007 of 200,

the cost savings that accrued to the Department of Corrections is around $5.8 million dollars per

year (in $2008). This figure is modest, and is in line with the limited use of electronic monitoring

by judges. However, an expansion of the use of electronic monitoring clearly has the potential

to produce larger savings. While it is difficult to speculate as to the number of eligible offenders

amongst those who served a prison sentence that would have been suitable to serve their sentence

under electronic monitoring, the fact that a significant proportion of judges never chose to use a

sentence of electronic monitoring suggests there is room for increased use of electronic monitoring

without compromising the suitability criteria, and hence success of the program.27

7 Discussion

The significant growth in prison populations and the attendant financial and social costs have led

to an increased interest in sentencing alternatives to imprisonment. A key concern when consid-

ering alternative punishment options is whether a move away from a heavy reliance on prison will

increase reoffending. In this paper we examine electronic monitoring as a front-end alternative

to imprisonment for non-serious and non-violent offences where the sentence length is up to 18

months. Our key objective is to evaluate whether the use of electronic monitoring rather than

prison impacts of reoffending, and if so whether it increases reoffending, or whether it decreases it.

In order to identify its causal effect, we address non-random selection of offenders into electronic

monitoring using an instrumental variables approach based on judge tendency for punishment type.

When considering all aged offenders, our findings suggest that the use of electronic monitoring as an

alternative to prison reduces the rate of reoffending within 24 months of free-time by 16 percentage

points, or 28% relative to the reoffending rate of those who are imprisoned. This reduction in

reoffending is sustained over a five year horizon. However, these impacts for all age offenders are

driven by those who are less than 30 years old at the time of their offense. For this group the

reduction in reoffending within 24 months is 43 percentage points, or 67%. Importantly, for this

younger group, who have a potentially longer future criminal career, the reduction in offender from

serving their sentence under electronic monitoring persists for 8 years. We also provide some rough

back-of-the-envelope calculations that suggest that each eligible offender who serves their sentence

under electronic monitoring rather than in prison produces a cost saving to the criminal justice

system of close to $30,000. This underlines the strong case for consideration of the use of electronic

27For example, the Auditor-General’s Report (2010) notes that over 6500 offenders were technically eligible forelectronic monitoring in NSW in 2008, with 1400 of these offenders committing traffic and motor vehicle regulatoryoffences. They also note that in 2008-09 only 35 out of 47 Local Courts with access to home detention referredoffenders for assessment.

25

monitoring as an alternative to imprisonment for less serious, non-violent offenders.

The interpretation of our findings as causal rests of several strong identifying assumptions. We

carefully investigate each of these assumptions, including the exclusion of the instrumental variable

from the structural equation, and monotonicity in the relationship between the instrument and the

endogenous outcome, in order to establish the internal validity of our approach. We make a case for

the external validity of our case study for jurisdictions considering the use of electronic monitoring

as a front-end alternative for imprisonment based on the broad set of offences and sentence lengths

eligible for electronic monitoring in the context studied, along with the program itself being mature

and operating at scale.

It is important to note that in the context we study, electronic monitoring both diverts of-

fenders from prison and provides individually tailored rehabilitation programs along with intense

supervision while the offender is living and working within the general community. With this in

mind, and given the strong case for both the internal and external validity of findings, the policy

implications are clear. They indicate that combining close monitoring and prescribed rehabilitation

for non-violent and non-serious offences, as occurs under electronic monitoring in the context we

study, has sustained crime reducing effects. Given that it reduces criminal justice costs as well as

reoffending, we conclude that electronic monitoring is a viable alternative to imprisonment.

26

References

Aizer, A., & Doyle, J.J. 2015. Juvenile Incarceration, Human Capital and Future Crime: Evidence fromRandomly-Assigned Judges. The Quarterly Journal of Economics, 130(2), 759–803.

Apel, R., Blokland, A., Nieuwbeertaand, P., & van Schellen, M. 2010. The impact of imprisonment onmarriage and divorce: A risk set matching approach. Journal of Quantitative Criminology, 26(2), 269300.

Auditor-General’s Report. 2010. Home Detention: Corrective Services NSW. Performance Audit.

Australian Bureau of Statistics. 2016. Census of Population and Housing: Australia Revealed.

Bayer, P., Hjalmarsson, R., & Pozen, D. 2009. Building Criminal Capital Behind Bars: Peer Effects inJuvenile Corrections. Quarterly Journal of Economics, 124(1), 105 –147.

Bhuller, M., Dahl, G., Lken, K., & Mogstad, M. 2016. Incarceration, Recidivism and Employment. NationalBureau of Economic Research Working Papers 22648.

Chalfin, A., & McCary, J. 2017. Criminal Deterrence: A Review of the Literature. Journal of EconomicLiterature, 55(1), 5–48.

DeMichele, M. 2014. Electronic Monitoring: It Is a Tool, Not a Silver Bullet. Criminology and Public Policy,13(3), 393–400.

DeVuono-Powell, S., Schweidler, C., Walters, A., & Zohrabi, A. 2015. Who pays? The true cost of incarcer-ation on families. Oakland, CA: Ella Baker Center.

Di Tella, R., & Schargrodsky, E. 2013. Criminal Recidivism after Prison and Electronic Monitoring. Journalof Political Economy, 121(1), 28 –73.

Dobbie, W., Goldin, J., & Yang, C. 2018. The Effects of Pre-Trial Detention on Conviction, Future Crime,and Employment: Evidence from Randomly Assigned Judges. American Economic Review, 108(2), 201–240.

Green, D.P., & Winik, D. 2010. Using Random Judge Assignments to Estimate the the Effects of Incarcer-ation and Probabtion on REcidivism Among Drug Offenders. Criminology, 48(2), 357–387.

Hawken, Angela, & Kleiman, Mark A. R. 2009. Managing Drug Involved Probationers with Swift and CertainSanctions: Evaluating Hawaiis HOPE. Washington, DC: National Criminal Justice Reference Service.

Heggie, K. 1999. Review of the NSW Homes Detention Scheme. Sydney: NSW Department of CorrectiveServices.

Henneguelle, A., Monnery, B., & Kensey, A. 2016. Better at Home than in Prison? The Effects of ElectronicMonitoring on Recidivism in France. The Journal of Law and Economics, 59(3), 629–667.

Hjalmarsson, R. 2009. Juvenile Jails: A Path to the Straight and Narrow or to Hardened Criminality?Journal of Law and Economics, 52(4), 779809.

Hucklesby, A., & Holdsworth, E. 2016. Electronic Monitoring in England and Wales. University of Leeds,UK: Centre for Criminal Justice Studies.

Kilmer, Beau, Nicosia, Nancy, Heaton, Paul, & Midgette, Greg. 2013. Efficacy of Frequent Monitoring withSwift, Certain, and Modest Sanctions for Violations: Insights from South Dakotas 24/7 Sobriety Project.American Journal of Public Health, 103(1), 37–43.

Kling, J. 2006. Incarceration Length, Employment, and Earnings. American Economic Review, 96(3),863–876.

27

Lopoo, L. M., & Western, B. 2005. Incarceration and the formation and stability of marital unions. Journalof Marriage and Family, 67(3), 72134.

Marie, O. 2009. The Best Ones Come Out First! Early Release from Prison and Recidivism A RegressionDiscontinuity Approach.

Mueller-Smith, M. 2015. The Criminal and Labor Market Impacts of Incarceration. University of MichiganWorking Paper.

Nagin, D. S., Cullen, F. T., & Jonson, C. L. 2009. Imprisonment and reoffending. Chap. 38, page 115200 of:Tonry, M (ed), Crime and justice: a review of research. Washington, D.C: University of Chicago Press.

Productivity Commission. 2010. Report on Government Services 2010,.

Renzema, M., & Mayo-Wilson, E. 2005. Can electronic monitoring reduce crime for moderate to high-riskoffenders? Journal of Experimental Criminology, 1(2), 215–237.

Studerus, K.J. 1997. Home Detention in NSW. Sydney: NSW Department of Corrective Services.

The Pew Charitable Trusts. 2016. Use of Electronic Offender-Tracking Devices Expands Sharply: Numberof monitored individuals more than doubled in 10 years. Tech. rept.

28

Tab

le1:

Dis

trib

uti

onof

Cri

me

Typ

eby

Man

ner

inw

hic

hth

eS

ente

nce

isS

erved

Fu

llS

am

ple

Est

imati

on

Sam

ple

crim

ety

pe

all

pri

son

elec

tron

icall

pri

son

elec

tron

icm

on

itori

ng

mon

itori

ng

%%

%%

%%

inju

ry6.9

97.4

82.3

97.2

58.0

52.1

1n

egli

gent

dri

vin

g3.2

13.1

33.9

64.1

34.0

24.8

2b

reak

&en

ter

9.6

510.2

83.7

79.9

610.9

63.6

1th

eft

24.6

926.2

310.2

421.1

122.9

29.5

4fr

aud

7.3

56.3

916.3

310.9

79.2

222.1

9p

ub

lic

ord

er6.5

67.0

51.9

55.8

16.4

12.0

1tr

affic

22.5

819.9

247.4

224.4

21.1

345.2

8G

ovt.

pro

ced

ure

s18.9

819.5

213.9

416.3

717.3

10.4

4

Tot

al100.0

090.3

49.6

6100.0

086.4

813.5

2

The

full

sam

ple

consi

sts

of

all

case

sb

efore

the

court

sb

etw

een

2000

and

2007

for

whic

hth

eoff

ender

isel

igib

leto

serv

eth

eir

sente

nce

under

elec

tronic

monit

ori

ng.

The

esti

mati

on

sam

ple

consi

sts

the

firs

tin

stance

that

each

off

ender

inth

efu

llsa

mple

isobse

rved

bef

ore

the

court

s.T

her

eare

16,4

75

case

sin

the

full

sam

ple

and

7,3

66

case

sin

the

esti

mati

on

sam

ple

.

29

Tab

le2:

Reo

ffen

din

gR

ates

by

Mon

ths

ofF

ree-

tim

ean

dM

ann

erin

Wh

ich

Sen

ten

ceis

Ser

ved

12

mth

24m

th36

mth

48m

th60

mth

72m

th84

mth

96m

th10

8m

th12

0m

th%

%%

%%

%%

%%

%

pri

son

47

5865

7073

7476

7881

84el

ectr

on

icm

onit

ori

ng

16

3037

4347

5153

5560

66

Ther

eare

7439

case

sfo

rre

off

endin

gat

12

month

sand

6769

case

sfo

rre

off

endin

gat

120

month

s.R

eoff

endin

gis

defi

ned

as

aco

urt

app

eara

nce

subse

quen

tto

the

index

off

ence

wit

hin

xm

onth

sof

free

-tim

e,w

her

efr

ee-t

ime

isti

me

inth

eco

mm

unit

y.

30

Tab

le3:

Fir

stS

tage

Est

imat

es:

Dep

end

ent

Var

iab

leis

Sen

ten

ced

Ser

ved

isE

M

Cou

rtX

Yea

rof

Fin

ali

zati

on

Ad

ds

for

Dem

ogra

ph

ics

Ad

ds

Cri

min

al

His

tory

Inte

ract

edF

ixed

Eff

ects

an

dC

rim

eT

yp

ean

dL

egal

Rep

rese

nta

tion

Rec

eive

EM

Rec

eive

EM

Rec

eive

EM

jud

gete

nd

ency

for

EM

0.6

1***

0.5

9***

0.5

9***

(0.0

8)

(0.0

8)

(0.0

8)

Fte

stof

excl

ud

edin

stru

men

tsF

(1,

63)

57.5

360.5

754.8

8(0

.00)

(0.0

0)

(0.0

0)

SW

un

der

iden

tifi

cati

onte

stC

hi-

sq.(

1)60.9

264.2

358.6

6(0

.00)

(0.0

0)

(0.0

0)

Ob

serv

atio

ns

7,3

66

7,3

66

7,3

66

Contr

ols

incl

ude

fully

inte

ract

edco

urt

and

yea

rof

finaliza

tion

fixed

effec

tsin

colu

mn

1.

Colu

mn

2adds

contr

ols

for

dem

ogra

phic

s(a

nin

dic

ato

rfo

rgen

der

ism

ale

,a

quadra

tic

inage)

and

indic

ato

rsfo

roff

ence

typ

eis

inju

ry,

neg

ligen

tdri

vin

g,

bre

ak

and

ente

r,th

eft,

fraud,

public

ord

er,

and

traffi

c.T

he

refe

rence

cate

gory

isvio

lati

ng

gov

ernm

ent

pro

cedure

s.C

olu

mn

3addit

ionally

incl

udes

contr

ols

for

the

num

ber

of

finalise

dco

urt

app

eara

nce

sw

ithin

5yea

rspri

or

toth

ecu

rren

tca

se,

an

indic

ato

rfo

rno

legal

repre

senta

tion

inth

ecu

rren

tca

se,

and

indic

ato

rsfo

rth

elo

cal

gov

ernm

ent

are

ain

whic

hth

eoff

ender

resi

des

.Sta

ndard

erro

rscl

ust

ered

on

the

court

are

rep

ort

edin

pare

nth

eses

.***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

31

Tab

le4:

Tes

tin

gfo

rR

and

omA

ssig

nm

ent

ofC

ases

toJu

dge

s

rece

ived

EM

jud

ge

ten

den

cym

ean

sd

age

-0.0

081***

0.0

001

31.3

89.6

5(0

.0020)

(0.0

007)

age

Squ

ared

0.0

001***

0.0

000

1077.7

5705.3

6(0

.0000)

(0.0

000)

gen

der

isM

ale

-0.1

048***

0.0

002

0.8

80.3

2(0

.0153)

(0.0

040)

no.

cou

rtap

pea

ran

ces

inth

ep

ast

5ye

ars

-0.0

077***

-0.0

006*

3.2

42.6

5(0

.0022)

(0.0

004)

no

lega

lre

pre

senta

tive

0.1

247***

0.0

004

0.0

80.2

8(0

.0202)

(0.0

050)

inju

ryoff

ence

-0.0

402**

-0.0

006

0.0

70.2

6(0

.0160)

(0.0

046)

neg

lige

nt

dri

vin

goff

ence

0.0

603**

-0.0

070

0.0

40.2

0(0

.0229)

(0.0

068)

bre

ak&

ente

roff

ence

0.0

016

0.0

020

0.1

00.3

0(0

.0111)

(0.0

031)

thef

toff

ence

-0.0

026

0.0

017

0.2

10.4

1(0

.0091)

(0.0

026)

frau

d0.1

084***

0.0

071

0.1

10.3

1(0

.0278)

(0.0

060)

pu

bli

cor

der

-0.0

081

0.0

002

0.0

60.2

3(0

.0147)

(0.0

046)

traffi

c0.1

373***

-0.0

003

0.2

40.4

3(0

.0214)

(0.0

026)

F-s

tat

31.2

91.4

0p

-val

ue

0.0

00.2

0

Ob

serv

atio

ns

7,3

66

7,3

66

R-s

qu

ared

0.2

678

0.0

358

Colu

mn

1re

port

sth

ere

sult

sfr

om

are

gre

ssio

nof

an

indic

ato

rfo

rre

ceiv

ing

ase

nte

nce

of

EM

on

the

off

ender

’scr

imin

al

and

per

sonal

chara

cter

isti

cs.

Colu

mn

2re

port

sth

ere

sult

sfr

om

are

gre

ssio

nof

the

judge

tenden

cyto

refe

roff

ender

s’fo

rass

essm

ent

for

EM

on

the

off

ender

scr

imin

al

and

per

sonal

chara

cter

isti

cs.

Coeffi

cien

tsand

standard

erro

rscl

ust

ered

on

at

the

court

level

(in

bra

cket

s).

The

F-s

tati

stic

and

its

p-v

alu

efo

rth

ete

stin

gth

enull

hyp

oth

esis

that

the

coeffi

cien

tsare

join

tly

zero

isre

port

edfo

rea

chre

gre

ssio

n.

The

mea

nand

standard

dev

iati

on

of

the

off

ender

s’cr

ime

and

chara

cter

isti

csare

rep

ort

edin

the

final

2co

lum

ns.

***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

32

Tab

le5:

Exam

inin

gM

onot

onic

ity

Base

lin

eIn

stru

ments

:F

irst

Sta

ge

Off

ence

Typ

eP

ast

Imp

riso

nm

ent

Age

traffi

cth

eft,

frau

d,

B&

Eoth

erp

ast

pri

son

no

past

pri

son

age<

30

age>

=30

jud

gete

nd

ency

for

EM

0.84

***

0.5

0***

0.4

4***

0.1

4*

0.7

3***

0.4

6***

0.6

8***

(-0.

26)

(0.1

0)

(0.0

9)

(0.0

8)

(0.1

1)

(0.0

7)

(0.1

4)

mea

nof

EM

0.25

10.1

14

0.0

78

0.0

35

0.1

75

0.1

03

0.1

72

Ob

serv

atio

ns

1,79

73,0

97

2,4

72

2,1

03

5,2

63

3,9

37

3,4

29

Revers

eIn

stru

ments

:F

irst

Sta

ge

Off

ence

Typ

eP

ast

Imp

riso

nm

ent

Age

traffi

cth

eft,

frau

d,

B&

Eoth

erp

ast

pri

son

no

past

pri

son

age<

30

age>

=30

jud

gete

nd

ency

for

EM

0.82

***

0.3

8**

0.2

4***

0.7

3***

0.6

0***

0.7

1***

0.7

6***

(0.2

9)(0

.13)

(0.0

6)

(0.2

1)

(0.0

9)

(0.1

4)

(0.1

6)

mea

nof

EM

0.25

10.1

08

0.0

78

0.0

36

0.1

76

0.1

02

0.1

72

Ob

serv

atio

ns

1,79

02,9

97

2,4

12

2,0

73

5,1

32

3,8

60

3,3

56

Coeffi

cien

tsand

standard

erro

rscl

ust

ered

on

at

the

court

level

(in

bra

cket

s).

The

top

panel

of

the

table

rep

ort

sth

efirs

tst

age

esti

mate

of

the

coeffi

cien

ton

the

inst

rum

enta

lva

riable

,w

her

eth

eIV

isco

nst

ruct

edusi

ng

the

full

sam

ple

and

wher

ees

tim

ati

on

of

the

firs

tst

age

isov

erth

esu

b-s

am

ple

defi

ned

by

the

colu

mn

hea

din

gs.

The

low

erpanel

rep

ort

sth

efirs

tst

age

esti

mate

of

the

coeffi

cien

ton

the

inst

rum

enta

lva

riable

,w

her

eth

eIV

isco

nst

ruct

edusi

ng

the

com

ple

men

tof

the

sub-s

am

ple

use

dfo

res

tim

ati

on

and

the

sub-s

am

ple

use

dfo

res

tim

ati

on

isdefi

ned

by

the

colu

mn

hea

din

gs.

***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

33

Tab

le6:

Bas

elin

eIn

stru

men

tal

Var

iab

leR

esu

lts:

Dep

end

ent

Var

iab

leis

Reo

ffen

din

gw

ith

in24

Mon

ths

ofF

reet

ime

Contr

ols

for

Cou

rtX

Yea

rof

Fin

aliz

atio

nA

dd

Con

trol

sfo

rD

emog

rap

hic

sA

dd

Con

trol

sfo

rC

rim

inal

His

tory

Inte

ract

edF

ixed

Eff

ects

and

Cri

me

Typ

ean

dL

egal

Rep

rese

nta

tion

Pr(

Reo

ffen

dw

ith

in24

mon

ths)

Pr(

Reo

ffen

dw

ith

in24

mon

ths)

Pr(

Reo

ffen

dw

ith

in24

mon

ths)

OL

Sse

nte

nce

isE

M-0

.25*

**-0

.16*

**-0

.14*

**(0

.02)

(0.0

2)(0

.02)

Red

uced

Form

jud

gete

nd

ency

for

EM

-0.1

7***

-0.1

2***

-0.0

9***

(0.0

6)(0

.05)

(0.0

4)IV se

nte

nce

isE

M-0

.27*

**-0

.21*

**-0

.16*

**(0

.08)

(0.0

7)(0

.07)

mea

n0.

580.

580.

58

Contr

ols

incl

ude

fully

inte

ract

edco

urt

and

yea

rof

finaliza

tion

fixed

effec

tsin

colu

mn

1.

Colu

mn

2adds

an

indic

ato

rfo

rgen

der

ism

ale

,a

quadra

tic

inage

and

indic

ato

rsfo

roff

ence

typ

eis

inju

ry,

neg

ligen

tdri

vin

g,

bre

ak

and

ente

r,th

eft,

fraud,

public

ord

er,

and

traffi

c.T

he

refe

rence

cate

gory

isvio

lati

ng

gov

ernm

ent

pro

cedure

s.C

olu

mn

3addit

ionally

incl

udes

contr

ols

for

the

num

ber

of

finalise

dco

urt

app

eara

nce

sw

ithin

5yea

rspri

or

toth

ecu

rren

tca

se,

an

indic

ato

rfo

rno

legal

repre

senta

tion

inth

ecu

rren

tca

se,

and

indic

ato

rsfo

rth

elo

cal

gov

ernm

ent

are

ain

whic

hth

eoff

ender

resi

des

.Sta

ndard

erro

rscl

ust

ered

on

the

court

are

rep

ort

edin

pare

nth

eses

.***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

34

Tab

le7:

Exam

inin

gth

eE

xcl

usi

onR

estr

icti

on

Fir

stS

tage

Rec

eive

EM

Rec

eive

EM

Rec

eive

EM

Sen

ten

ceL

ength

jud

gete

nd

ency

for

EM

0.5

9***

0.5

6***

0.5

6***

-0.1

9(0

.08)

(0.0

8)

(.08)

(0.6

9)

jud

gete

nd

ency

for

Sen

ten

ceL

engt

h–

0.0

10.0

10.7

7***

(0.0

03)

(0.0

03)

(0.0

6)

F-s

tat

(in

stru

men

t)54.8

851.1

9–

–(0

.000)

(0.0

00)

SW

Ch

i-sq

(1)

un

der

ID–

–55.3

1145.3

(0.0

00)

(0.0

00)

Red

uced

Form

Pr(

Reo

ffen

dw

ith

in24

month

s)P

r(R

eoff

end

wit

hin

24

month

s)P

r(R

eoff

end

wit

hin

24

month

s)

jud

gete

nd

ency

for

EM

-0.0

9**

-0.1

0**

-0.1

0**

(0.0

4)

(0.0

4)

(0.0

4)

jud

gete

nd

ency

for

Sen

ten

ceL

engt

h0.0

01

0.0

01

(0.0

04

)(0

.004)

IVP

r(R

eoff

end

wit

hin

24

month

s)P

r(R

eoff

end

wit

hin

24

month

s)P

r(R

eoff

end

wit

hin

24

month

s)

rece

ived

EM

-0.1

6***

-0.1

7***

-0.1

7***

(0.0

7)

(0.0

6)

(0.0

6)

sente

nce

len

gth

––

0.0

03

(0.0

05)

jud

gete

nd

ency

sente

nce

len

gth

–0.0

02

–(0

.004)

mea

n0.5

80.5

80.5

8

All

spec

ifica

tions

contr

ol

for

inte

ract

edco

urt

and

yea

rof

finaliza

tion

fixed

effec

ts,

an

indic

ato

rfo

rgen

der

ism

ale

,a

quadra

tic

inage,

indic

ato

rsfo

roff

ence

typ

eis

inju

ry,

neg

ligen

tdri

vin

g,

bre

ak

and

ente

r,th

eft,

fraud,

public

ord

er,

and

traffi

c(r

efer

ence

cate

gory

isvio

lati

ng

gov

ernm

ent

pro

cedure

s),

the

num

ber

of

finalise

dco

urt

app

eara

nce

sw

ithin

5yea

rspri

or

toth

ecu

rren

tca

se,

an

indic

ato

rfo

rno

legal

repre

senta

tion

inth

ecu

rren

tca

se,

and

indic

ato

rsfo

rth

elo

cal

gov

ernm

ent

are

ain

whic

hth

eoff

ender

resi

des

.Sta

ndard

erro

rscl

ust

ered

on

the

court

are

rep

ort

edin

pare

nth

eses

.***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

35

Tab

le8:

Rob

ust

nes

s

Reo

ffen

dw

ith

in24

mon

ths

Cou

rtap

pea

ran

ces

wit

hin

24m

onth

s

adju

stE

Mn

op

riso

nn

op

riso

nag

e<30

age>

=30

lnn

o.ap

pea

ran

ces

no.

ofap

pea

ran

ces

free

tim

ein

pri

or5

yrs

&n

otre

man

ded

sente

nce

isE

M-0

.19*

*-0

.20*

**-0

.13*

*-0

.43*

*0.

02-0

.52*

-1.1

0(0

.06)

(0.0

7)(0

.07)

(0.1

8)(0

.14)

(0.3

0)(0

.61)

Ob

serv

atio

ns

7,3

285,

263

2,93

33,

644

3,72

269

9969

99

Contr

ols

incl

ude

fully

inte

ract

edco

urt

and

yea

rof

finaliza

tion

fixed

effec

ts,

dem

ogra

phic

s(a

nin

dic

ato

rfo

rgen

der

ism

ale

,and

aquadra

tic

inage)

and

indic

ato

rsfo

roff

ence

typ

eis

inju

ry,

neg

ligen

tdri

vin

g,

bre

ak

and

ente

r,th

eft,

fraud,

public

ord

er,

and

traffi

c,re

fere

nce

cate

gory

isvio

lati

ng

gov

ernm

ent

pro

cedure

s),

crim

inal

his

tory

(the

num

ber

of

finalise

dco

urt

app

eara

nce

sw

ithin

5yea

rspri

or

toth

ecu

rren

tca

se),

an

indic

ato

rfo

rno

legal

repre

senta

tion

inth

ecu

rren

tca

se,

and

indic

ato

rsfo

rth

elo

cal

gov

ernm

ent

are

ain

whic

hth

eoff

ender

resi

des

.Sta

ndard

erro

rscl

ust

ered

on

the

court

are

rep

ort

edin

pare

nth

eses

.***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

36

Tab

le9:

Inst

rum

enta

lV

aria

ble

Est

imat

es:

Imp

act

ofR

ecei

vin

ga

Sen

ten

ceof

Ele

ctro

nic

Mon

itor

ing

by

Du

rati

onof

Fre

etim

e

Pro

bab

ilit

yof

reoff

en

din

gw

ith

in12

mth

24

mth

36

mth

48

mth

60

mth

72

mth

84

mth

96

mth

108

mth

120

mth

A.

All

Ages

serv

edse

nte

nce

as

EM

-0.2

0*

-0.1

6**

-0.1

4-0

.14*

-0.1

5*-0

.12

-0.1

1-0

.10

-0.0

6-0

.03

(0.1

0)(0

.07)

(0.0

9)(0

.08)

(0.0

9)(0

.08)

(0.0

8)(0

.07)

(0.0

7)(0

.07)

mea

n0.

470.5

80.

650.

700.

730.

740.

760.

780.

810.

84nu

mb

erof

case

s7,4

597,

366

7,32

77,

309

7,28

87,

271

7,25

37,

221

7,01

36,

769

B.

Less

than

30

serv

edse

nte

nce

as

EM

-0.4

1*

-0.4

3**

-0.3

9-0

.32

-0.3

5-0

.33

-0.3

7*-0

.31*

-0.2

2-0

.22

(0.2

2)(0

.18)

(0.2

6)(0

.22)

(0.2

4)(0

.22)

(0.2

0)(0

.19)

(0.1

8)(0

.17)

mea

n0.

510.6

40.

710.

750.

770.

790.

800.

810.

840.

86nu

mb

erof

case

s3,6

833,

644

3,62

43,

618

3,61

53,

609

3,60

23,

589

3,51

13,

443

B.

At

least

30

serv

edse

nte

nce

as

EM

-0.0

90.0

20.

030.

02-0

.01

0.05

0.07

0.07

0.09

0.12

(0.1

4)(0

.14)

(0.1

5)(0

.16)

(0.1

5)(0

.14)

(0.1

4)(0

.12)

(0.1

3)(0

.14)

mea

n0.

410.5

20.

600.

650.

680.

700.

720.

740.

770.

81O

bse

rvati

on

s3,7

76

3,7

223,

703

3,69

13,

673

3,66

23,

651

3,63

23,

502

3,32

6

All

spec

ifica

tions

incl

ude

the

full

set

of

contr

ols

.Sta

ndard

erro

rscl

ust

ered

on

the

court

are

rep

ort

edin

pare

nth

eses

.***

p<

0.0

1,

**

p<

0.0

5,

*p<

0.1

37

Table 10: Seriousness of Reoffence Type

prison (%) EM (%)all offenders given reoffend all offenders given reoffend

did not reoffend 42 – 70 –offence not serious 42 72 22 73offence serious 17 29 8 27

Based on author calculations. Serious offences are: homicide, injury, sexual assault, negligent acts, abductions, androbbery. Less serious offences are: break and enter, theft, fraud, drugs, weapons, property damage, public order,traffic, government procedures, miscellaneous.

Table 11: Reoffending and Sentencing

prison (%) EM (%)all offenders given reoffend all offenders given reoffend

did not reoffend 42 70unsupervised 25 43 15 50supervised, not custody 8 13 5 16custodial sentence 25 44 10 35

Based on author calculations. Custodial sentences: prison, electronic monitoring, periodic detention and intensivecorrection order. Supervised, noncustodial: suspended sentence with supervision, community service order, bondwith supervision, probation with supervision. Unsupervised: suspended sentence without supervision, bond withoutsupervision, probation without supervision, fine, non-criminal sentence, bond without conviction, no conviction.

Table 12: Net Savings per Offender Sentenced to EM

Savings per EM sentence

savings from reduced cost for current offence 25200court cases averted 449custodial sentences averted 3603

Total 29252

As averted court appearances and custodial sentences occur any time from the start of free-time up to 24 months offreetime. We assume averted court appearances and custodial sentences occur at the end of the 24 month offreetime period, and discount these savings at the rate of 5%.

38

Fig

ure

1:

Ten

den

cyof

Ju

dge

sto

Ref

erfo

rA

sses

smen

tfo

rE

lect

ron

icM

onit

orin

g

0.01.02.03.04.05Fraction

.01

.02

.03

.04

.06

.07

.08

.09

.11

.12

.13

.14

.16

.17

.18

.19

.21

.22

.23

.24

0.0

5.1

.15

.2.2

5pr

ob o

f jud

ge fa

ced

refe

rring

to e

lect

roni

c m

onito

ring

asse

ssm

ent

Est

imat

ion

sam

ple

con

sist

sof

7,36

6ca

ses

fin

aliz

edb

etw

een

2000

an

d2007.

Th

eh

isto

gra

msh

ows

the

den

sity

of

jud

ge

ten

den

cyfo

rre

ferr

ing

case

sfo

ras

sess

men

tfo

rsu

itab

ilit

yfo

rth

eel

ectr

onic

mon

itori

ng

pro

gra

mad

just

edfo

rco

urt

an

dye

ar

of

fin

ali

zati

on

inte

ract

edfi

xed

effec

ts.

Th

eto

pan

db

otto

m5%

ofob

serv

atio

ns

hav

eb

een

excl

ud

ed.

39